Research On Color Structured Light Based Three-dimensional Measurement Registration And Point Cloud Object Recognition Method

Three-dimensional visual inspection technology has greatly promoted the transformation and upgrading of traditional manufacturing to intelligent automation.How to perform efficient and stable 3D measurement,3D recognition and localization is a major challenge in the field of 3D vision research.Aiming at the problem that the speed and accuracy of the structured light 3D measurement in the field of industrial inspection are currently restricted,this paper proposed a new composite color coding based 3D measuring method.Under the premise of ensuring a certain measurement accuracy,the number of projections was greatly reduced and the measurement speed was improved effectively.Aiming at the problems of registration misalignment and low algorithm efficiency of state-of-the-art multi-perspective point cloud automatic registration methods,this paper proposed a multiple matching constraints principle based coarse-fine combination point cloud registration method,which greatly reduced the wrong feature matching point pairs and improved multi-perspective point cloud registration accuracy and efficiency.In addition,for the problem of low accuracy of the existing 3D object recognition method in complex scenes,a weighting features based machine learning recognition scheme and a weighting point pairs feature based recognition method were studied.The weighting factor improved the accuracy of identification algorithm.Finally,a fast and high-precision 3D measurement splicing and recognition system was integrated and developed.The main contents of this article are summarized as follows:1.In-depth research on three-dimensional vision technologies such as structured light three-dimensional measurement,multi-perspective point cloud registration and point cloud recognition for complex scenes.Summarize and contrast the existing methods.Point out the current difficult problems in the field of 3D vision and establish the research focus of this article.2.For the contradictory problems between high-speed and high-precision of the structured light 3D measuring technology,a new color-coding based method combining sinusoidal-coding,stair-phase-coding and gray-coding was proposed,which can not only ensure the 3D measurement precision,but also greatly improve the measurement efficiency with less projection patterns.Meanwhile,to improve the accuracy of the 3D measuring system,the Caspi color response model and the seventh-order polynomial intensity response model were utilized to compensate the crosstalk and nonlinear error,respectively.3.Aiming at the problems of low efficiency and low registration accuracy of multiperspective 3D point cloud registration,a multiple matching-constraints based 3D point cloud registration method was studied.Considering the direct impact of the mismatching feature matching pairs on the accuracy of point cloud registration results,innovatively combining the principles of reciprocity,geometric consistency and random sampling consistency to accurately select the feature matching pairs of the point cloud to be registered,which can avoid the registration calculation process of the incorrect matching point pairs and improve registration efficiency and accuracy.4.In view of the low robustness of the existing 3D recognition methods for complex scenes,the weighting features based point cloud recognition methods were studied.The traditional machine learning based recognition method combining the global and local point cloud features and the point pair features based recognition method were thoroughly studied.Considering the influence of point cloud noise and density,the weighting calculation is performed in the feature obtaining process,and the advantages and disadvantages of the two schemes before and after weighting were analyzed and compared,so as to select the optimal solution for developing the relevant functional modules of the system.5.Based on the above theoretical and methodological research,C++,the application development framework Qt,the image processing library Open CV,the point cloud processing library PCL and the matrix operation library Eigen are used to implement various algorithmic functional modules,and a set of fast and robust 3D measurement and recognition systems for objects was developed.The 3D measuring system was developed with the measuring range of 300 x 250 x 30 mm,the measuring accuracy of 200?m,and the measurement time of 1.3s.Finally,the functions of the 3D vision system and the effectiveness of relevant algorithms were verified through the experiment of 3D measurement and recognition of the objects.